The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill architecture is susceptible to indirect prompt injection through external data sources.
Ingestion points: User-provided prompts in submitGeneration and the request body in the express callback handler in SKILL.md.
Boundary markers: The implementation lacks boundary markers or specific instructions to the agent to disregard instructions within the processed data.
Capability inventory: The skill utilizes filesystem writes via fs.writeFileSync and sharp, and network capabilities via fetch.
Sanitization: There is no evidence of input validation or sanitization for prompts or metadata before they are processed or written to disk.
[DATA_EXFILTRATION]: The skill contains logic to send data to a user-specified callbackUrl via fetch. While this is part of the intended callback pattern, it represents a network operation to a non-whitelisted domain that could be leveraged to exfiltrate information processed within the pipeline.

ideogram-webhooks-events